In [458]:
import numpy as np
import cv2
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
from sklearn.preprocessing import StandardScaler
import glob
import random
from random import shuffle
from skimage.feature import hog
from sklearn.model_selection import train_test_split
from scipy.ndimage.measurements import label
import os
import pickle
from collections import deque
from feature_extraction_classification import *
Look at some HOG transforms¶
Gray Scale Visualization of HOG (using default parameters )¶
In [5]:
rand_non_car_filename = random.choice(glob.glob('non-vehicles/*/*.png'))
rand_car_filename = random.choice(glob.glob('vehicles/*/*.png'))
nocar_img = mpimg.imread(rand_non_car_filename)
grayscale_nocar = convert_color(nocar_img,'GRAY')
car_img = mpimg.imread(rand_car_filename)
grayscale_car = convert_color(car_img,'GRAY')
hog_noncar_features, hog_noncar_image = get_hog_features( grayscale_nocar,vis=True)
hog_car_features, hog_car_image = get_hog_features(grayscale_car,vis=True)
# Plot the examples
f, (ax1,ax2) = plt.subplots(1, 2, figsize=(10, 5))
f.tight_layout()
ax1.imshow(nocar_img)
ax1.set_title('Non Car Image')
ax2.imshow(hog_noncar_image, cmap='gray')
ax2.set_title('Hog Visualization of NonCar image')
plt.savefig('submission_images/Car_hog_visualization.png',bbox_inches='tight')
f, (ax1,ax2) = plt.subplots(1, 2, figsize=(10, 5))
f.tight_layout()
ax1.imshow(car_img)
ax1.set_title('Car Image')
ax2.imshow(hog_car_image, cmap='gray')
ax2.set_title('Hog Visualization of NonCar image')
plt.savefig('submission_images/Non-car_hog_visualization.png',bbox_inches='tight')
plt.show()
Visualization of color feature¶
In [14]:
rand_non_car_filename_2 = random.choice(glob.glob('non-vehicles/*/*.png'))
rand_car_filename_2 = random.choice(glob.glob('vehicles/*/*.png'))
nocar_img_2 = mpimg.imread(rand_non_car_filename_2)
car_img_2 = mpimg.imread(rand_car_filename_2)
color_noncar_conv_img = convert_color(nocar_img_2,'HSV')
color_noncar = bin_spatial(color_noncar_conv_img)
color_car_conv_img = convert_color(car_img_2,'HSV')
color_car = bin_spatial(color_car_conv_img)
# Plot the examples
f, (ax1,ax2) = plt.subplots(1, 2, figsize=(10, 5))
f.tight_layout()
ax1.imshow(nocar_img_2)
ax1.set_title('Non Car Image')
ax2.plot(color_noncar)
ax2.set_title('Color Visualization of NonCar image')
plt.savefig('submission_images/Car_color_visualization.png',bbox_inches='tight')
f, (ax1,ax2) = plt.subplots(1, 2, figsize=(10, 5))
f.tight_layout()
ax1.imshow(car_img_2)
ax1.set_title('Car Image')
ax2.plot(color_car)
ax2.set_title('Color Visualization of Car image')
plt.savefig('submission_images/non-Car_color_visualization.png',bbox_inches='tight')
plt.show()
In [9]:
rand_non_car_filename_3 = random.choice(glob.glob('non-vehicles/*/*.png'))
rand_car_filename_3 = random.choice(glob.glob('vehicles/*/*.png'))
nocar_img_3 = mpimg.imread(rand_non_car_filename_3)
car_img_3 = mpimg.imread(rand_car_filename_3)
color_hist_noncar = bin_spatial(nocar_img_3)
color_hist_car = bin_spatial(car_img_3)
# Plot the examples
f, (ax1,ax2) = plt.subplots(1, 2, figsize=(10, 5))
f.tight_layout()
ax1.imshow(nocar_img_3)
ax1.set_title('Non Car Image')
ax2.plot(color_hist_noncar)
ax2.set_title('Color histogram of NonCar image')
plt.savefig('submission_images/non-Car_color_hist_visualization.png',bbox_inches='tight')
f, (ax1,ax2) = plt.subplots(1, 2, figsize=(10, 5))
f.tight_layout()
ax1.imshow(car_img_3)
ax1.set_title('Car Image')
ax2.plot(color_hist_car)
ax2.set_title('Color histogram of Car image')
plt.savefig('submission_images/Car_color_hist_visualzation.png',bbox_inches='tight')
plt.show()
Build generic Classifier¶
In [291]:
list_car_img_filename = glob.glob('vehicles/*/*.png')
list_noncar_img_filename = glob.glob('non-vehicles/*/*.png')
list_cars_images = []
list_noncar_images = []
for im in list_car_img_filename:
list_cars_images.append(im)
for im in list_noncar_img_filename:
list_noncar_images.append(im)
print("Total number of Car images ={0}, and non Car images = {1}".format(len(list_cars_images), len(list_noncar_images)))
sample_size = len(list_noncar_images)
#sample_size = 500
shuffle(list_cars_images)
shuffle(list_noncar_images)
list_cars_image_sample = list_cars_images[0:sample_size]
list_noncars_image_sample = list_noncar_images[0:sample_size]
In [15]:
scaler, classifier, bin_spatial_acc= build_classifier_with_feature_extraction(list_cars_image_sample,
list_noncars_image_sample,
colorspace='RGB2YCrCb',
bin_spatial_feat=True,
color_hist_feat=True,
hog_feature=False,
hog_features_lists ={}
)
print("Accuracy based off color feature={0}".format(bin_spatial_acc))
In [16]:
hog_features_lists ={}
hog_features_lists ['hog_channels'] = 'ALL'
hog_features_lists ['orient'] = 9
hog_features_lists ['pixels_per_cell'] = 8
hog_features_lists ['cells_per_block'] = 2
shuffle(list_cars_image_sample)
shuffle(list_noncars_image_sample)
color_scaler, classifier, hog_only_acc= build_classifier_with_feature_extraction(list_cars_image_sample,
list_noncars_image_sample,
colorspace='RGB2YCrCb',
bin_spatial_feat=False,
color_hist_feat=False,
hog_feature=True,
hog_features_lists = hog_features_lists
)
print("Accuracy based off hog only feature={0}".format(hog_only_acc))
Analyze classifier accuracy based on color distribution¶
In [23]:
color_list = ['HLS', 'HSV', 'RGB2YCrCb', 'BGR2YCrCb']
for color in color_list:
hog_features_lists ={}
hog_features_lists ['hog_channels'] = 'ALL'
hog_features_lists ['orient'] = 9
hog_features_lists ['pixels_per_cell'] = 8
hog_features_lists ['cells_per_block'] = 2
shuffle(list_cars_image_sample)
shuffle(list_noncars_image_sample)
color_scaler, classifier, acc= build_classifier_with_feature_extraction(list_cars_image_sample,
list_noncars_image_sample,
colorspace=color,
bin_spatial_feat=False,
color_hist_feat=False,
hog_feature=True,
hog_features_lists = hog_features_lists
)
print("Color = {0}, Accuracy={1}".format(color, acc))
Analyze classifier accuracy based on color histogram¶
Find best hog channel keeping rest of them fixed.¶
Color = 'RGB2YCrCb', orientation = 9, pixels_per_block =2, cells_per_block=2 Hog channels can be 0 , 1, 2 or ALL
In [24]:
hog_channel_list = [0,1,2,'ALL']
for hog_channel in hog_channel_list:
hog_features_lists ={}
hog_features_lists ['hog_channels'] = hog_channel
hog_features_lists ['orient'] = 9
hog_features_lists ['pixels_per_cell'] = 8
hog_features_lists ['cells_per_block'] = 2
shuffle(list_cars_image_sample)
shuffle(list_noncars_image_sample)
scaler, classifier, acc_channel= build_classifier_with_feature_extraction(list_cars_image_sample,
list_noncars_image_sample,
colorspace='BGR2YCrCb',
bin_spatial_feat=False,
color_hist_feat=False,
hog_feature=True,
hog_features_lists = hog_features_lists
)
print("HOG Channel = {0}, Accuracy={1}".format(hog_channel, acc_channel))
Try different orientation¶
In [25]:
orient_exp_list = [5,9,11,15]
for orient_exp in orient_exp_list:
hog_features_lists ={}
hog_features_lists ['hog_channels'] = 'ALL'
hog_features_lists ['orient'] = orient_exp
hog_features_lists ['pixels_per_cell'] = 2
hog_features_lists ['cells_per_block'] = 2
file_name ='orient_feature_acc' + str(orient_exp) + '.p'
if os.path.isfile(file_name):
acc_orient = pickle.load( open( file_name, "rb" ) )
else:
scaler, classifier, acc_orient= build_classifier_with_feature_extraction(list_cars_image_sample,
list_noncars_image_sample,
colorspace='BGR2YCrCb',
bin_spatial_feat=False,
color_hist_feat=False,
hog_feature=True,
hog_features_lists = hog_features_lists
)
pickle.dump(acc_orient, open( file_name, "wb" ) )
print("Num of Orientation = {0}, Accuracy={1}".format(orient_exp, acc_orient))
In [26]:
pixels_list = [4,6,8,12]
for pixels_val in pixels_list:
hog_features_lists ={}
hog_features_lists ['hog_channels'] = 'ALL'
hog_features_lists ['orient'] = 11
hog_features_lists ['pixels_per_cell'] = pixels_val
hog_features_lists ['cells_per_block'] = 2
shuffle(list_cars_image_sample)
shuffle(list_noncars_image_sample)
file_name ='pixel_feature_acc' + str(pixels_val) + '.p'
if os.path.isfile(file_name):
acc_pixel = pickle.load( open( file_name, "rb" ) )
else:
scaler_pixels, classifier, acc_pixel= build_classifier_with_feature_extraction(list_cars_image_sample,
list_noncars_image_sample,
colorspace='BGR2YCrCb',
bin_spatial_feat=False,
color_hist_feat=False,
hog_feature=True,
hog_features_lists = hog_features_lists
)
pickle.dump(acc_pixel, open( file_name, "wb" ) )
print("Pixels per cell = {0}, Accuracy={1}".format(pixels_val, acc_pixel))
In [27]:
cells_list = [2, 4, 8]
for cells_val in cells_list:
hog_features_lists ={}
hog_features_lists ['hog_channels'] = 'ALL'
hog_features_lists ['orient'] = 11
hog_features_lists ['pixels_per_cell'] = 8
hog_features_lists ['cells_per_block'] = cells_val
shuffle(list_cars_image_sample)
shuffle(list_noncars_image_sample)
file_name ='cells_list_feature_acc' + str(pixels_val) + '.p'
if os.path.isfile(file_name):
acc_cells = pickle.load( open( file_name, "rb" ))
else:
scaler, classifier, acc_cells = build_classifier_with_feature_extraction(list_cars_image_sample,
list_noncars_image_sample,
colorspace='BGR2YCrCb',
bin_spatial_feat=False,
color_hist_feat=False,
hog_feature=True,
hog_features_lists = hog_features_lists
)
pickle.dump(acc_cells, open( file_name, "wb" ) )
print("Cells Per block = {0}, Accuracy={1}".format(cells_val, acc_cells))
In [46]:
hog_features_lists ={}
hog_features_lists ['hog_channels'] = 'ALL'
hog_features_lists ['orient'] = 11
hog_features_lists ['pixels_per_cell'] = 8
hog_features_lists ['cells_per_block'] = 2
shuffle(list_cars_image_sample)
shuffle(list_noncars_image_sample)
if os.path.isfile('all_feat_accuracy.p'):
scaler_all_feat = pickle.load( open('all_feat_scaler.p', "rb" ) )
class_all_feat = pickle.load( open('all_feat_classifier.p', "rb" ) )
acc_all_features = pickle.load( open('all_feat_accuracy.p', "rb" ) )
else:
scaler_all_feat, class_all_feat, acc_all_features= build_classifier_with_feature_extraction(list_cars_image_sample,
list_noncars_image_sample,
colorspace='BGR2YCrCb',
bin_spatial_feat = True,
color_hist_feat = True,
hog_feature= True,
hog_features_lists=hog_features_lists)
pickle.dump(scaler_all_feat, open('all_feat_scaler.p', "wb" ) )
pickle.dump(class_all_feat, open('all_feat_classifier.p', "wb" ) )
pickle.dump(acc_all_features, open('all_feat_accuracy.p', "wb" ) )
print("All feature extraction={0}".format(acc_all_features))
In [292]:
hog_features_lists ={}
hog_features_lists ['hog_channels'] = 'ALL'
hog_features_lists ['orient'] = 11
hog_features_lists ['pixels_per_cell'] = 8
hog_features_lists ['cells_per_block'] = 2
shuffle(list_cars_image_sample)
shuffle(list_noncars_image_sample)
if os.path.isfile('all_feat_scaler_rgb.p'):
scaler_all_feat_rgb = pickle.load( open('all_feat_scaler_rgb.p', "rb" ) )
class_all_feat_rgb = pickle.load( open('all_feat_classifier_rgb.p', "rb" ) )
acc_all_features_rgb = pickle.load( open('all_feat_accuracy_rgb.p', "rb" ) )
else:
scaler_all_feat_rgb, class_all_feat_rgb, acc_all_features_rgb= build_classifier_with_feature_extraction(list_cars_image_sample,
list_noncars_image_sample,
colorspace='RGB2YCrCb',
bin_spatial_feat = True,
color_hist_feat = True,
hog_feature= True,
hog_features_lists=hog_features_lists)
pickle.dump(scaler_all_feat_rgb, open('all_feat_scaler_rgb.p', "wb" ) )
pickle.dump(class_all_feat_rgb, open('all_feat_classifier_rgb.p', "wb" ) )
pickle.dump(acc_all_features_rgb, open('all_feat_classifier_rgb.p', "wb" ) )
print("All feature extraction with RGB2YCrCb={0}".format(acc_all_features_rgb))
This is the best accuracy achieved. Using this scaler(scaler_all_feat_rgb) and classifier for the rest of experiment (class_all_feat_rgb)¶
Start with Windowing with brute-force method
In [30]:
test_images = glob.glob('test_images/*.png')
for window_test_img in test_images:
slid_win_test_img = mpimg.imread(window_test_img)
windows = slide_window(slid_win_test_img, x_start_stop=(0, slid_win_test_img.shape[1]),
y_start_stop=(400, slid_win_test_img.shape[0]),
xy_window=(128, 128), xy_overlap=(0.5, 0.5))
window_img = draw_boxes(slid_win_test_img, windows, color=(0, 0, 255), thick=6)
f, (ax1,ax2) = plt.subplots(1, 2, figsize=(20, 9))
f.tight_layout()
ax1.imshow(slid_win_test_img)
ax1.set_title('Original Image')
ax2.imshow(window_img)
ax2.set_title('Sliding Window')
plt.show()
In [50]:
test_images = glob.glob('test_images/*.png')
hog_features_lists ={}
hog_features_lists ['hog_channels'] = 'ALL'
hog_features_lists ['orient'] = 11
hog_features_lists ['pixels_per_cell'] = 8
hog_features_lists ['cells_per_block'] = 2
for idx, test_image in enumerate(test_images):
search_window_test_img = mpimg.imread(test_image)
draw_img = np.copy(search_window_test_img)
wind = slide_window(draw_img, x_start_stop=(0, search_window_test_img.shape[1]),
y_start_stop=(400, search_window_test_img.shape[0]),
xy_window=(128, 128), xy_overlap=(0.5, 0.5))
hot_windows = hot_windows = search_windows(draw_img, wind, class_hogbin_feat, scaler_hogbin_feat,
colorspace='BGR2YCrCb',
bin_spatial_feat = True,
color_hist_feat = False,
hog_feature= True,
hog_features_lists=hog_features_lists)
hot_window_img = draw_boxes(draw_img, hot_windows, color=(0, 0, 255), thick=6)
f, (ax1,ax2) = plt.subplots(1, 2, figsize=(20, 9))
f.tight_layout()
ax1.imshow(search_window_test_img)
ax1.set_title('Original Test Image')
ax2.imshow(hot_window_img)
ax2.set_title('Car Bboxes')
file_name = 'submission_images/search_window_output_'+ str(idx) +'.png'
plt.savefig(file_name,bbox_inches='tight')
plt.show()
Optimized extraction of features using find_cars()¶
In [51]:
test_images = glob.glob('test_images/*.jpg')
for idx, test_img in enumerate(test_images):
opt_image = cv2.imread(test_img)
opt_image = cv2.cvtColor(opt_image, cv2.COLOR_BGR2RGB)
out_img = find_cars(opt_image, ystart=400, ystop=656,
scale=1.5, svc=class_all_feat_rgb,
scaler_obj=scaler_all_feat_rgb,
orient=11,
pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32)
plt.imshow(out_img)
file_name = 'submission_images/find_car_output_image_'+ str(idx) +'.png'
plt.savefig(file_name,bbox_inches='tight')
plt.show()
In [280]:
#Improve the bounding box accuracry by applying heatmap and threshold
In [293]:
test_images = glob.glob('test_images/*.jpg')
for idx, test_img in enumerate(test_images):
img = cv2.imread(test_img)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
box_list = find_cars_bbox(img, ystart=400, ystop=656,
scale=1.5, svc=class_all_feat,
X_scaler=scaler_all_feat,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32)
out_img = find_cars(img, ystart=400, ystop=656,
scale=1.5, svc=class_all_feat,
scaler_obj=scaler_all_feat,
orient=11,
pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32)
# Add heat to each box in box list
heat_init= np.zeros_like(img[:,:,0]).astype(np.float)
heat = add_heat(heat_init,box_list)
# Apply threshold to help remove false positives
heat = apply_threshold(heat,1)
# Visualize the heatmap when displaying
heatmap = np.clip(heat, 0, 255)
# Find final boxes from heatmap using label function
from scipy.ndimage.measurements import label
labels = label(heatmap)
draw_img = draw_labeled_bboxes(np.copy(img), labels)
f, (ax1,ax2, ax3,ax4) = plt.subplots(1, 4, figsize=(20, 9))
f.tight_layout()
ax1.imshow(img)
ax1.set_title('Original Test Image')
ax2.imshow(out_img)
ax2.set_title('Bbox image')
ax3.imshow(heatmap)
ax3.set_title('heat map')
ax4.imshow(draw_img)
ax4.set_title('After applying thresholding')
file_name = 'submission_images/find_car_false_postives'+ str(idx) +'.png'
plt.savefig(file_name,bbox_inches='tight')
plt.show()
Explore Different search areas to find and remove false positives and recognize true positives.¶
In [457]:
test_images = glob.glob('test_images/*.jpg')
for idx, test_img in enumerate(test_images):
img = cv2.imread(test_img)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
bbox_list =[]
bbox_list.append(find_cars_bbox(img, ystart=400, ystop=500,
scale=1.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=430, ystop=530,
scale=1.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=400, ystop=520,
scale=1.5, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=420, ystop=550,
scale=1.5, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=440, ystop=560,
scale=2.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=460, ystop=580,
scale=2.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=400, ystop=620,
scale=3.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=440, ystop=660,
scale=3.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
rectangles = [item for sublist in bbox_list for item in sublist]
test_img_rects = draw_boxes(img, rectangles, color=(0, 0, 255), thick=2)
heat_init= np.zeros_like(img[:,:,0]).astype(np.float)
heat = add_heat(heat_init,rectangles)
# Apply threshold to help remove false positives
if len(bbox_list) > 10:
heat = apply_threshold(heat,10)
else :
heat = apply_threshold(heat,1)
# Visualize the heatmap when displaying
heatmap = np.clip(heat, 0, 255)
# Find final boxes from heatmap using label function
from scipy.ndimage.measurements import label
labels = label(heatmap)
draw_img = draw_labeled_bboxes(np.copy(img), labels)
f, (ax1,ax2, ax3,ax4) = plt.subplots(1, 4, figsize=(20, 9))
f.tight_layout()
ax1.imshow(img)
ax1.set_title('Original Test Image')
ax2.imshow(test_img_rects)
ax2.set_title('Bbox image')
ax3.imshow(heatmap)
ax3.set_title('heat map')
ax4.imshow(draw_img)
ax4.set_title('After applying thresholding')
# file_name = 'submission_images/find_car_heatmap_scaling_experiment_'+ str(idx) +'.png'
# plt.savefig(file_name,bbox_inches='tight')
plt.show()
Video processing pipeline¶
In [471]:
class Vehicle_Detect():
def __init__(self):
self.cache = deque(maxlen=10)
In [477]:
def process_vid_frm(img):
bbox_list = []
bbox_list.append(find_cars_bbox(img, ystart=400, ystop=500,
scale=1.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=430, ystop=530,
scale=1.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=400, ystop=520,
scale=1.5, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=420, ystop=550,
scale=1.5, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=440, ystop=560,
scale=2.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=460, ystop=580,
scale=2.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=400, ystop=620,
scale=3.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
bbox_list.append(find_cars_bbox(img, ystart=440, ystop=660,
scale=3.0, svc=class_all_feat_rgb,
X_scaler=scaler_all_feat_rgb,
orient=11, pix_per_cell=8,
cell_per_block=2, spatial_size=(32,32), hist_bins=32))
heat = np.zeros_like(img[:,:,0]).astype(np.float)
# Add heat to each box in box list
for b in bbox_list:
if (len(b) != 0):
heat = add_heat(heat,b)
det.cache.append(heat)
heat = np.sum(det.cache, axis=0)
heat = apply_threshold(heat,15)
# Visualize the heatmap when displaying
heatmap = np.clip(heat, 0, 255)
# Find final boxes from heatmap using label function
labels = label(heatmap)
draw_img = draw_labeled_bboxes(np.copy(img), labels)
return draw_img
In [480]:
from moviepy.editor import VideoFileClip
from IPython.display import HTML
import os
det = Vehicle_Detect()
output = 'output_video/vehicleDetection.mp4'
clip1 = VideoFileClip("project_video.mp4")
white_clip = clip1.fl_image(process_vid_frm) #NOTE: this function expects color images!!
%time white_clip.write_videofile(output, audio=False)
In [481]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(output))
Out[481]: